Your AI Strategy Is Tuition, And Waiting Makes It More Expensive

Summary

• The biggest gains in AI adoption didn’t come from picking the right license or platform. They came from getting curious people talking to each other.
• Your most valuable AI champions might not be your most technical people — they’re often the ones who understand your actual workflows best.
• Give AI agents one narrow job each instead of one big one. It’s the difference between something you can monitor and something you can’t.
• Compliance isn’t a reason to wait on an AI strategy. Regulated businesses that already know their data are often better positioned to move, not worse.
• The cost of getting your team using these tools is about as low as it will ever be. Every month you wait is a month you’re paying more to learn the same lesson.

To close out the AI Starter Series, Kyle Etter, CIT’s President and CEO, sat down to talk strategy — the big-picture thinking that ties together everything the series covered on policy, security, and getting started. One question kicked it off: if you were starting your AI rollout over today, knowing what you know now, what would you do differently?

His answer wasn’t about tools. It was about people.

The Real Multiplier Wasn’t a License. It Was a Conversation.

Ask Kyle what actually moved the needle, and he doesn’t point to a platform decision or a cost analysis. He points to getting curious people into a room and letting them talk.

“There’s such a hunger for everybody to try to understand,” he says, and that hunger is where the real traction came from — not procurement. The pattern was simple: find the people with natural curiosity, get them talking, and watch what happens. One person tries something, another sees it and thinks, “I didn’t think about doing that,” and suddenly they’re adapting it for their own work. That’s the snowball. That’s the multiplier.

The mistake he owns: being too slow to hand out licenses. Staring down a $30-per-user Copilot bill across 150 people adds up fast, and it’s easy to hesitate. Looking back, he wishes he’d gotten licenses into certain hands sooner. As he puts it, “you don’t know what you don’t know” — hindsight’s always 20/20. The lesson stuck: community and engagement matter more than which platform you pick or where you start automating.

Your Champions Might Not Be Who You’d Guess

Here’s a twist Kyle didn’t expect: at CIT, the earliest, most enthusiastic adopters weren’t the most technical people on the team. Some genuinely skilled field engineers spent months watching from the sidelines, saying things like “I don’t really need AI, I don’t like AI.” Six months later, those same people started coming to him saying, “I’ve been listening, I’ve been joining, and I’m trying to figure this out.”

The shift didn’t come from the most advanced technical staff — it came from people who deeply understand the day-to-day workflows and processes of the business, technical or not. Give those people room to show what they know, and they’ll often surprise you with who steps up.

Give Every Agent One Job, Not the Whole Business

Naturally, the conversation turned to a harder question: when is it safe to let AI run on its own, and when does a human need to stay in the loop?

For Kyle, it comes down to risk assessment. Before you let an agent operate fully on its own, you need the confidence that its job is narrow and tightly scoped — something you can actually watch, measure, and control. The wider the scope, the wider the surface area for something to go wrong. His advice: you’re better off running several agents each with a tightly limited role than one agent trying to do everything.

That means building in the boring-but-critical stuff up front — audit trails, tripwires, alerts for when something isn’t behaving the way you expected. And before any of that goes near real business risk, you need to actually test and understand how these agents behave. Start narrow, prove it works, then expand.

Compliance Isn’t the Roadblock You Think It Is

For businesses in healthcare, government, or finance, “AI strategy” often gets filed under “someday, once we’ve figured out compliance.” Kyle pushes back on that framing.

It starts with picking the right vendor — one whose licensing actually aligns with your compliance requirements. A common misstep: people default to free, personal-tier AI tools that were never built for business use, and in the public sector, many of those tools simply aren’t authorized to begin with. From there, it’s about knowing your data: what you have, where it lives, what needs protecting — then building rules for how AI can and can’t touch it. Business-grade licensing from providers like Anthropic, OpenAI, Microsoft, or Google is table stakes here, and tools like Microsoft Purview make data classification and sensitivity labeling genuinely achievable.

Here’s the part that surprised even Kyle: this same problem already existed with people, not just AI. Businesses tend to trust their own employees more, on the assumption that “they read the policy.” Most people, if you’re honest, couldn’t recite that policy back to you. Classifying your data pays off twice — it lets your AI tools operate safely, and it lets you catch a person accidentally emailing sensitive information externally or dropping it into a Teams chat, the exact same way you’d catch an agent doing it.

And on that note, there’s an odd upside to agents: once you tell one “you can’t forward sensitive files outside the organization,” it actually won’t — every time. People don’t always maintain that same discipline. Which means a regulated business that already knows its data well may have an easier path into AI than a company with looser data habits, not a harder one.

The Tuition Only Gets More Expensive From Here

Asked what separates the businesses moving on AI strategy today from the ones still waiting, Kyle doesn’t hedge: if you’re in the “waiting” camp, it’s worth rethinking that.

His advice for getting moving: get the right licenses into the right hands, identify your early adopters, give them a way to collaborate, and consider bringing in outside help — a consulting partner like CIT — to mentor the process. Put a real AI committee in place, write down your policy (the series covered this in an earlier episode), and start classifying your data if you haven’t already.

The financial case is straightforward. These subscriptions are cheap right now, and Kyle expects that to change as adoption grows. Get your people using the tools while the cost is low, and you’re compounding value while it’s still a bargain. Wait, and you’ll eventually pay a much higher tuition to learn the exact same lessons — just later, and at a disadvantage to everyone who didn’t wait.

Missed any part of the AI Starter Series? The full strategy webinar — and the rest of the series — is linked here.